C++ - Avoid allocating memory with dqsample_int

[TylerSagendorf]

Hello, I am doing some Monte Carlo simulation—using dqset_seed and dqsample_int in C++—and I was wondering is there a way to avoid allocating memory for each sample if I want to simply overwrite the contents of an existing Rcpp::IntegerVector? Below is an example of how I currently have the code set up:

// Example values
const int n = 15000;
const int m =  1000;
const int n_samples = 1000000;

// Set seed and stream
dqrng::dqset_seed(IntegerVector::create(0), IntegerVector::create(99));

IntegerVector idx(m);

for (int i = 0; i < n_samples; ++i) {
  // I would like to update the values of idx, 
  // rather than copying the result of dqsample_int
  idx = dqrng::dqsample_int(n, m);

  // Use idx to select elements from different vectors
  // ...
}

I am still not that familiar with C++, so I apologize if there is an obvious solution.

[rstub]

Any reason why you want to do this? I would expect that return value optimization leads to one vector with 1000 elements being allocated per loop. In principle, R has a "copy on write" policy, that means one should not change an existing vector directly. I don't think this is relevant here, but I am hesitant to offer something like that in the public API.

[TylerSagendorf]

Sorry for the (very late) reply. My original idea was to see if there was a way to avoid additional memory allocations by somehow modifying the elements of the idx vector directly, since garbage collection was being triggered quite often (according to profvis::profvis()), but now I am not sure that it would meaningfully reduce the total runtime.